An additive approach to transform-domain information hiding and optimum detection structure

This paper presents an additive approach to transform-domain information hiding and the performance analysis for images and video. The watermark embedding method is designed to satisfy the perceptual constraints and improve the detectability as well as the information embedding rate. The statistical behaviors of subband coefficients are modeled by the generalized Gaussian distribution. The structure of the optimum detection is built and the performance of the exact asymptotic detection is evaluated using large deviation theory. Our approach can not only achieve good transparency but also precisely control the detection errors. It can be applied to watermarking, authentication, fingerprinting, and steganography.